maria-discuss team mailing list archive

[Jeff Dyke <jeff.dyke@xxxxxxxxx>]

I asked here a while ago in here about my inability to make physical
mariabackups viable in a new server, it would eventually fail a restart ( i
had a test that failed that led me to look into this).  This was in one AWS
environment(VPC).  In the end there were no disk problems, no corrupt
tables, no read issues with aria or innodbcheck as well as an online check
with mysqlcheck.  And as i said in the earlier email, it works perfect in
staging.

Some of you may remember that.  Anyway, to my satisfaction, i have solved
that problem, and its more scorched earth solution than anything, happy to
elaborate after.

No I have two very stable slaves set in strict gtid mode and I want to add
them to the cluster that i am having an issue with, eventually retiring old
nodes.  I have brought up new cluster nodes on so many occasions, but never
transitioned from a replica and i don't trust the output of the current
cluster SSTs, but it may work now, with my new setups.  I'll test that and
ask this at the same time.

I can layout a base plan for what i would do, but i'm asking for folks that
have had to deal with this to compare my notes with.

- All servers in the same VPC, spread across AZs, all IAM roles/policies
fully in place.
- These are mariadb/galera nodes, not members of an RDS cluster.  I had
needs that RDS didn't have when i built out this layout, in case you're
curious or want to say move to RDS.

Thanks if you have pointers.  This is all mariadb 10.4.12/13.

If i can tie this one up, i can release a bunch of simple set of bash
scripts but make it safe to run on clusters, regardless if you're testing
or provisioning, even as a pipiline test.

Thanks again,